Investigating and Evaluating Deep Learning Approaches for Addressing Class Imbalances in Threat Detection from X-ray Baggage Images

Abstract

The detection of threat items in baggage X-ray scans is one of the most essential yet most challenging tasks, where deep learning has been showing promising improvements. The latter comes as an alternative to the otherwise cumbersome, slow, and error-ridden expert-based manual analysis of X-ray scans. Currently, vast amounts of baggage cross air, sea, and land ports continuously, rendering said advancements highly important. However, much of the proposed solutions suffer from the problem of data imbalance, where the presence of threatened items is low, which affects the performance of networks. This project aims to find approaches to address the problem of data imbalance and utilize it in instance segmentation, detection, and classification frameworks to enhance performance.

Date of Award	Dec 2022
Original language	American English
Supervisor	Naoufel Werghi (Supervisor)